4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 * Directory code for lustre client.
42 #include <linux/pagemap.h>
44 #include <linux/version.h>
45 #include <linux/user_namespace.h>
46 #ifdef HAVE_UIDGID_HEADER
47 # include <linux/uidgid.h>
49 #include <asm/uaccess.h>
50 #include <linux/buffer_head.h> // for wait_on_buffer
51 #include <linux/pagevec.h>
53 #define DEBUG_SUBSYSTEM S_LLITE
55 #include <lustre/lustre_idl.h>
57 #include <obd_support.h>
58 #include <obd_class.h>
59 #include <lustre_ioctl.h>
60 #include <lustre_lib.h>
61 #include <lustre_dlm.h>
62 #include <lustre_fid.h>
63 #include <lustre_kernelcomm.h>
64 #include <lustre_swab.h>
66 #include "llite_internal.h"
69 * (new) readdir implementation overview.
71 * Original lustre readdir implementation cached exact copy of raw directory
72 * pages on the client. These pages were indexed in client page cache by
73 * logical offset in the directory file. This design, while very simple and
74 * intuitive had some inherent problems:
76 * . it implies that byte offset to the directory entry serves as a
77 * telldir(3)/seekdir(3) cookie, but that offset is not stable: in
78 * ext3/htree directory entries may move due to splits, and more
81 * . it is incompatible with the design of split directories for cmd3,
82 * that assumes that names are distributed across nodes based on their
83 * hash, and so readdir should be done in hash order.
85 * New readdir implementation does readdir in hash order, and uses hash of a
86 * file name as a telldir/seekdir cookie. This led to number of complications:
88 * . hash is not unique, so it cannot be used to index cached directory
89 * pages on the client (note, that it requires a whole pageful of hash
90 * collided entries to cause two pages to have identical hashes);
92 * . hash is not unique, so it cannot, strictly speaking, be used as an
93 * entry cookie. ext3/htree has the same problem and lustre implementation
94 * mimics their solution: seekdir(hash) positions directory at the first
95 * entry with the given hash.
101 * Client caches directory pages using hash of the first entry as an index. As
102 * noted above hash is not unique, so this solution doesn't work as is:
103 * special processing is needed for "page hash chains" (i.e., sequences of
104 * pages filled with entries all having the same hash value).
106 * First, such chains have to be detected. To this end, server returns to the
107 * client the hash of the first entry on the page next to one returned. When
108 * client detects that this hash is the same as hash of the first entry on the
109 * returned page, page hash collision has to be handled. Pages in the
110 * hash chain, except first one, are termed "overflow pages".
112 * Solution to index uniqueness problem is to not cache overflow
113 * pages. Instead, when page hash collision is detected, all overflow pages
114 * from emerging chain are immediately requested from the server and placed in
115 * a special data structure (struct ll_dir_chain). This data structure is used
116 * by ll_readdir() to process entries from overflow pages. When readdir
117 * invocation finishes, overflow pages are discarded. If page hash collision
118 * chain weren't completely processed, next call to readdir will again detect
119 * page hash collision, again read overflow pages in, process next portion of
120 * entries and again discard the pages. This is not as wasteful as it looks,
121 * because, given reasonable hash, page hash collisions are extremely rare.
123 * 1. directory positioning
125 * When seekdir(hash) is called, original
136 * identification of and access to overflow pages
140 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
141 * a header lu_dirpage which describes the start/end hash, and whether this
142 * page is empty (contains no dir entry) or hash collide with next page.
143 * After client receives reply, several pages will be integrated into dir page
144 * in PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
145 * lu_dirpage for this integrated page will be adjusted. See
146 * mdc_adjust_dirpages().
149 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
150 __u64 offset, struct ll_dir_chain *chain)
152 struct md_callback cb_op;
156 cb_op.md_blocking_ast = ll_md_blocking_ast;
157 rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
164 void ll_release_page(struct inode *inode, struct page *page,
169 /* Always remove the page for striped dir, because the page is
170 * built from temporarily in LMV layer */
171 if (inode != NULL && S_ISDIR(inode->i_mode) &&
172 ll_i2info(inode)->lli_lsm_md != NULL) {
179 if (likely(page->mapping != NULL))
180 truncate_complete_page(page->mapping, page);
183 page_cache_release(page);
187 * return IF_* type for given lu_dirent entry.
188 * IF_* flag shld be converted to particular OS file type in
189 * platform llite module.
191 static u16 ll_dirent_type_get(struct lu_dirent *ent)
194 struct luda_type *lt;
197 if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
198 const unsigned align = sizeof(struct luda_type) - 1;
200 len = le16_to_cpu(ent->lde_namelen);
201 len = (len + align) & ~align;
202 lt = (void *)ent->lde_name + len;
203 type = IFTODT(le16_to_cpu(lt->lt_type));
209 #ifdef HAVE_DIR_CONTEXT
210 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
211 struct dir_context *ctx)
214 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
215 void *cookie, filldir_t filldir)
218 struct ll_sb_info *sbi = ll_i2sbi(inode);
220 bool is_api32 = ll_need_32bit_api(sbi);
221 bool is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
223 struct ll_dir_chain chain;
228 ll_dir_chain_init(&chain);
230 page = ll_get_dir_page(inode, op_data, pos, &chain);
232 while (rc == 0 && !done) {
233 struct lu_dirpage *dp;
234 struct lu_dirent *ent;
243 hash = MDS_DIR_END_OFF;
244 dp = page_address(page);
245 for (ent = lu_dirent_start(dp); ent != NULL && !done;
246 ent = lu_dirent_next(ent)) {
253 hash = le64_to_cpu(ent->lde_hash);
256 * Skip until we find target hash
261 namelen = le16_to_cpu(ent->lde_namelen);
268 if (is_api32 && is_hash64)
272 fid_le_to_cpu(&fid, &ent->lde_fid);
273 ino = cl_fid_build_ino(&fid, is_api32);
274 type = ll_dirent_type_get(ent);
275 /* For 'll_nfs_get_name_filldir()', it will try
276 * to access the 'ent' through its 'lde_name',
277 * so the parameter 'name' for 'filldir()' must
278 * be part of the 'ent'. */
279 #ifdef HAVE_DIR_CONTEXT
281 done = !dir_emit(ctx, ent->lde_name, namelen, ino,
284 done = filldir(cookie, ent->lde_name, namelen, lhash,
291 ll_release_page(inode, page, false);
295 next = le64_to_cpu(dp->ldp_hash_end);
297 if (pos == MDS_DIR_END_OFF) {
299 * End of directory reached.
302 ll_release_page(inode, page, false);
305 * Normal case: continue to the next
308 ll_release_page(inode, page,
309 le32_to_cpu(dp->ldp_flags) &
312 page = ll_get_dir_page(inode, op_data, pos,
316 #ifdef HAVE_DIR_CONTEXT
321 ll_dir_chain_fini(&chain);
325 #ifdef HAVE_DIR_CONTEXT
326 static int ll_iterate(struct file *filp, struct dir_context *ctx)
328 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
331 struct inode *inode = file_inode(filp);
332 struct ll_file_data *lfd = LUSTRE_FPRIVATE(filp);
333 struct ll_sb_info *sbi = ll_i2sbi(inode);
334 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
335 int api32 = ll_need_32bit_api(sbi);
336 struct md_op_data *op_data;
346 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
347 "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
348 inode, (unsigned long)pos, i_size_read(inode), api32);
350 if (pos == MDS_DIR_END_OFF)
356 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
357 LUSTRE_OPC_ANY, inode);
359 GOTO(out, rc = PTR_ERR(op_data));
361 if (unlikely(op_data->op_mea1 != NULL)) {
362 /* This is only needed for striped dir to fill ..,
363 * see lmv_read_entry */
364 if (file_dentry(filp)->d_parent != NULL &&
365 file_dentry(filp)->d_parent->d_inode != NULL) {
366 __u64 ibits = MDS_INODELOCK_UPDATE;
367 struct inode *parent =
368 file_dentry(filp)->d_parent->d_inode;
370 if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
371 op_data->op_fid3 = *ll_inode2fid(parent);
374 /* If it can not find in cache, do lookup .. on the master
376 if (fid_is_zero(&op_data->op_fid3)) {
377 rc = ll_dir_get_parent_fid(inode, &op_data->op_fid3);
379 ll_finish_md_op_data(op_data);
384 op_data->op_max_pages = sbi->ll_md_brw_pages;
385 #ifdef HAVE_DIR_CONTEXT
387 rc = ll_dir_read(inode, &pos, op_data, ctx);
390 rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
395 if (pos == MDS_DIR_END_OFF) {
397 pos = LL_DIR_END_OFF_32BIT;
399 pos = LL_DIR_END_OFF;
404 #ifdef HAVE_DIR_CONTEXT
409 ll_finish_md_op_data(op_data);
410 filp->f_version = inode->i_version;
414 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
419 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
421 struct mgs_send_param *msp;
428 strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
429 rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
430 sizeof(struct mgs_send_param), msp, NULL);
432 CERROR("Failed to set parameter: %d\n", rc);
439 * Create striped directory with specified stripe(@lump)
441 * param[in]parent the parent of the directory.
442 * param[in]lump the specified stripes.
443 * param[in]dirname the name of the directory.
444 * param[in]mode the specified mode of the directory.
446 * retval =0 if striped directory is being created successfully.
447 * <0 if the creation is failed.
449 static int ll_dir_setdirstripe(struct inode *parent, struct lmv_user_md *lump,
450 const char *dirname, umode_t mode)
452 struct ptlrpc_request *request = NULL;
453 struct md_op_data *op_data;
454 struct ll_sb_info *sbi = ll_i2sbi(parent);
455 struct inode *inode = NULL;
456 struct dentry dentry;
460 if (unlikely(lump->lum_magic != LMV_USER_MAGIC))
463 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s"
464 "stripe_offset %d, stripe_count: %u\n",
465 PFID(ll_inode2fid(parent)), parent, dirname,
466 (int)lump->lum_stripe_offset, lump->lum_stripe_count);
468 if (lump->lum_stripe_count > 1 &&
469 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
472 if (IS_DEADDIR(parent) &&
473 !OBD_FAIL_CHECK(OBD_FAIL_LLITE_NO_CHECK_DEAD))
476 if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
477 lustre_swab_lmv_user_md(lump);
479 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
480 mode &= ~current_umask();
481 mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
482 op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
483 strlen(dirname), mode, LUSTRE_OPC_MKDIR,
486 GOTO(err_exit, err = PTR_ERR(op_data));
488 op_data->op_cli_flags |= CLI_SET_MEA;
489 err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
490 from_kuid(&init_user_ns, current_fsuid()),
491 from_kgid(&init_user_ns, current_fsgid()),
492 cfs_curproc_cap_pack(), 0, &request);
493 ll_finish_md_op_data(op_data);
497 err = ll_prep_inode(&inode, request, parent->i_sb, NULL);
501 memset(&dentry, 0, sizeof(dentry));
502 dentry.d_inode = inode;
504 err = ll_init_security(&dentry, inode, parent);
510 ptlrpc_req_finished(request);
514 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
517 struct ll_sb_info *sbi = ll_i2sbi(inode);
518 struct md_op_data *op_data;
519 struct ptlrpc_request *req = NULL;
521 struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
522 struct obd_device *mgc = lsi->lsi_mgc;
528 * This is coming from userspace, so should be in
529 * local endian. But the MDS would like it in little
530 * endian, so we swab it before we send it.
532 switch (lump->lmm_magic) {
533 case LOV_USER_MAGIC_V1: {
534 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
535 lustre_swab_lov_user_md_v1(lump);
536 lum_size = sizeof(struct lov_user_md_v1);
539 case LOV_USER_MAGIC_V3: {
540 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
541 lustre_swab_lov_user_md_v3(
542 (struct lov_user_md_v3 *)lump);
543 lum_size = sizeof(struct lov_user_md_v3);
546 case LMV_USER_MAGIC: {
547 if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
548 lustre_swab_lmv_user_md(
549 (struct lmv_user_md *)lump);
550 lum_size = sizeof(struct lmv_user_md);
554 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
555 " %#08x != %#08x nor %#08x\n",
556 lump->lmm_magic, LOV_USER_MAGIC_V1,
562 lum_size = sizeof(struct lov_user_md_v1);
565 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
566 LUSTRE_OPC_ANY, NULL);
568 RETURN(PTR_ERR(op_data));
570 /* swabbing is done in lov_setstripe() on server side */
571 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
572 ll_finish_md_op_data(op_data);
573 ptlrpc_req_finished(req);
575 if (rc != -EPERM && rc != -EACCES)
576 CERROR("mdc_setattr fails: rc = %d\n", rc);
579 /* In the following we use the fact that LOV_USER_MAGIC_V1 and
580 LOV_USER_MAGIC_V3 have the same initial fields so we do not
581 need the make the distiction between the 2 versions */
582 if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
586 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
588 GOTO(end, rc = -ENOMEM);
591 /* Get fsname and assume devname to be -MDT0000. */
592 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
593 strcat(buf, "-MDT0000.lov");
596 /* Set root stripesize */
597 sprintf(buf, ".stripesize=%u",
598 lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
599 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
603 /* Set root stripecount */
604 sprintf(buf, ".stripecount=%hd",
605 lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
606 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
610 /* Set root stripeoffset */
611 sprintf(buf, ".stripeoffset=%hd",
612 lump ? le16_to_cpu(lump->lmm_stripe_offset) :
613 (typeof(lump->lmm_stripe_offset))(-1));
614 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
618 OBD_FREE(param, MGS_PARAM_MAXLEN);
624 * This function will be used to get default LOV/LMV/Default LMV
625 * @valid will be used to indicate which stripe it will retrieve
626 * OBD_MD_MEA LMV stripe EA
627 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
628 * otherwise Default LOV EA.
629 * Each time, it can only retrieve 1 stripe EA
631 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
632 struct ptlrpc_request **request, u64 valid)
634 struct ll_sb_info *sbi = ll_i2sbi(inode);
635 struct mdt_body *body;
636 struct lov_mds_md *lmm = NULL;
637 struct ptlrpc_request *req = NULL;
639 struct md_op_data *op_data;
642 rc = ll_get_default_mdsize(sbi, &lmm_size);
646 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
647 0, lmm_size, LUSTRE_OPC_ANY,
650 RETURN(PTR_ERR(op_data));
652 op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
653 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
654 ll_finish_md_op_data(op_data);
656 CDEBUG(D_INFO, "md_getattr failed on inode "
657 DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
661 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
662 LASSERT(body != NULL);
664 lmm_size = body->mbo_eadatasize;
666 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
668 GOTO(out, rc = -ENODATA);
671 lmm = req_capsule_server_sized_get(&req->rq_pill,
672 &RMF_MDT_MD, lmm_size);
673 LASSERT(lmm != NULL);
676 * This is coming from the MDS, so is probably in
677 * little endian. We convert it to host endian before
678 * passing it to userspace.
680 /* We don't swab objects for directories */
681 switch (le32_to_cpu(lmm->lmm_magic)) {
683 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
684 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
687 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
688 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
691 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
692 lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
695 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
696 lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
699 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
704 *plmm_size = lmm_size;
709 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
711 struct md_op_data *op_data;
716 OBD_ALLOC_PTR(op_data);
720 op_data->op_flags |= MF_GET_MDT_IDX;
721 op_data->op_fid1 = *fid;
722 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
723 mdt_index = op_data->op_mds;
724 OBD_FREE_PTR(op_data);
732 * Get MDT index for the inode.
734 int ll_get_mdt_idx(struct inode *inode)
736 return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
740 * Generic handler to do any pre-copy work.
742 * It sends a first hsm_progress (with extent length == 0) to coordinator as a
743 * first information for it that real work has started.
745 * Moreover, for a ARCHIVE request, it will sample the file data version and
746 * store it in \a copy.
748 * \return 0 on success.
750 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
752 struct ll_sb_info *sbi = ll_s2sbi(sb);
753 struct hsm_progress_kernel hpk;
758 /* Forge a hsm_progress based on data from copy. */
759 hpk.hpk_fid = copy->hc_hai.hai_fid;
760 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
761 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
762 hpk.hpk_extent.length = 0;
765 hpk.hpk_data_version = 0;
768 /* For archive request, we need to read the current file version. */
769 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
771 __u64 data_version = 0;
773 /* Get inode for this fid */
774 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
776 hpk.hpk_flags |= HP_FLAG_RETRY;
777 /* hpk_errval is >= 0 */
778 hpk.hpk_errval = -PTR_ERR(inode);
779 GOTO(progress, rc = PTR_ERR(inode));
782 /* Read current file data version */
783 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
786 CDEBUG(D_HSM, "Could not read file data version of "
787 DFID" (rc = %d). Archive request ("
788 LPX64") could not be done.\n",
789 PFID(©->hc_hai.hai_fid), rc,
790 copy->hc_hai.hai_cookie);
791 hpk.hpk_flags |= HP_FLAG_RETRY;
792 /* hpk_errval must be >= 0 */
793 hpk.hpk_errval = -rc;
797 /* Store in the hsm_copy for later copytool use.
798 * Always modified even if no lsm. */
799 copy->hc_data_version = data_version;
803 /* On error, the request should be considered as completed */
804 if (hpk.hpk_errval > 0)
805 hpk.hpk_flags |= HP_FLAG_COMPLETED;
807 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
810 /* Return first error */
811 RETURN(rc != 0 ? rc : rc2);
815 * Generic handler to do any post-copy work.
817 * It will send the last hsm_progress update to coordinator to inform it
818 * that copy is finished and whether it was successful or not.
821 * - for ARCHIVE request, it will sample the file data version and compare it
822 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
823 * will be considered as failed.
824 * - for RESTORE request, it will sample the file data version and send it to
825 * coordinator which is useful if the file was imported as 'released'.
827 * \return 0 on success.
829 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
831 struct ll_sb_info *sbi = ll_s2sbi(sb);
832 struct hsm_progress_kernel hpk;
837 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
838 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
839 * initialized if copy_end was called with copy == NULL.
842 /* Forge a hsm_progress based on data from copy. */
843 hpk.hpk_fid = copy->hc_hai.hai_fid;
844 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
845 hpk.hpk_extent = copy->hc_hai.hai_extent;
846 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
847 hpk.hpk_errval = copy->hc_errval;
848 hpk.hpk_data_version = 0;
850 /* For archive request, we need to check the file data was not changed.
852 * For restore request, we need to send the file data version, this is
853 * useful when the file was created using hsm_import.
855 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
856 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
857 (copy->hc_errval == 0)) {
859 __u64 data_version = 0;
861 /* Get lsm for this fid */
862 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
864 hpk.hpk_flags |= HP_FLAG_RETRY;
865 /* hpk_errval must be >= 0 */
866 hpk.hpk_errval = -PTR_ERR(inode);
867 GOTO(progress, rc = PTR_ERR(inode));
870 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
873 CDEBUG(D_HSM, "Could not read file data version. "
874 "Request could not be confirmed.\n");
875 if (hpk.hpk_errval == 0)
876 hpk.hpk_errval = -rc;
880 /* Store in the hsm_copy for later copytool use.
881 * Always modified even if no lsm. */
882 hpk.hpk_data_version = data_version;
884 /* File could have been stripped during archiving, so we need
885 * to check anyway. */
886 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
887 (copy->hc_data_version != data_version)) {
888 CDEBUG(D_HSM, "File data version mismatched. "
889 "File content was changed during archiving. "
890 DFID", start:"LPX64" current:"LPX64"\n",
891 PFID(©->hc_hai.hai_fid),
892 copy->hc_data_version, data_version);
893 /* File was changed, send error to cdt. Do not ask for
894 * retry because if a file is modified frequently,
895 * the cdt will loop on retried archive requests.
896 * The policy engine will ask for a new archive later
897 * when the file will not be modified for some tunable
899 hpk.hpk_flags &= ~HP_FLAG_RETRY;
901 /* hpk_errval must be >= 0 */
902 hpk.hpk_errval = -rc;
909 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
912 /* Return first error */
913 RETURN(rc != 0 ? rc : rc2);
917 static int copy_and_ioctl(int cmd, struct obd_export *exp,
918 const void __user *data, size_t size)
923 OBD_ALLOC(copy, size);
927 if (copy_from_user(copy, data, size)) {
932 rc = obd_iocontrol(cmd, exp, size, copy, NULL);
934 OBD_FREE(copy, size);
939 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
941 int cmd = qctl->qc_cmd;
942 int type = qctl->qc_type;
943 int id = qctl->qc_id;
944 int valid = qctl->qc_valid;
951 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
955 if (((type == USRQUOTA &&
956 !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
958 !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
959 (!cfs_capable(CFS_CAP_SYS_ADMIN)))
965 CERROR("unsupported quotactl op: %#x\n", cmd);
969 if (valid != QC_GENERAL) {
970 if (cmd == Q_GETINFO)
971 qctl->qc_cmd = Q_GETOINFO;
972 else if (cmd == Q_GETQUOTA)
973 qctl->qc_cmd = Q_GETOQUOTA;
979 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
980 sizeof(*qctl), qctl, NULL);
983 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
984 sizeof(*qctl), qctl, NULL);
987 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
988 sizeof(*qctl), qctl, NULL);
990 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
992 sizeof(*qctl), qctl, NULL);
1004 struct obd_quotactl *oqctl;
1006 OBD_ALLOC_PTR(oqctl);
1010 QCTL_COPY(oqctl, qctl);
1011 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1013 OBD_FREE_PTR(oqctl);
1016 /* If QIF_SPACE is not set, client should collect the
1017 * space usage from OSSs by itself */
1018 if (cmd == Q_GETQUOTA &&
1019 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1020 !oqctl->qc_dqblk.dqb_curspace) {
1021 struct obd_quotactl *oqctl_tmp;
1023 OBD_ALLOC_PTR(oqctl_tmp);
1024 if (oqctl_tmp == NULL)
1025 GOTO(out, rc = -ENOMEM);
1027 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1028 oqctl_tmp->qc_id = oqctl->qc_id;
1029 oqctl_tmp->qc_type = oqctl->qc_type;
1031 /* collect space usage from OSTs */
1032 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1033 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1034 if (!rc || rc == -EREMOTEIO) {
1035 oqctl->qc_dqblk.dqb_curspace =
1036 oqctl_tmp->qc_dqblk.dqb_curspace;
1037 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1040 /* collect space & inode usage from MDTs */
1041 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1042 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1043 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1044 if (!rc || rc == -EREMOTEIO) {
1045 oqctl->qc_dqblk.dqb_curspace +=
1046 oqctl_tmp->qc_dqblk.dqb_curspace;
1047 oqctl->qc_dqblk.dqb_curinodes =
1048 oqctl_tmp->qc_dqblk.dqb_curinodes;
1049 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1051 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1054 OBD_FREE_PTR(oqctl_tmp);
1057 QCTL_COPY(qctl, oqctl);
1058 OBD_FREE_PTR(oqctl);
1064 /* This function tries to get a single name component,
1065 * to send to the server. No actual path traversal involved,
1066 * so we limit to NAME_MAX */
1067 static char *ll_getname(const char __user *filename)
1072 OBD_ALLOC(tmp, NAME_MAX + 1);
1075 return ERR_PTR(-ENOMEM);
1077 len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1080 else if (len > NAME_MAX)
1081 ret = -ENAMETOOLONG;
1084 OBD_FREE(tmp, NAME_MAX + 1);
1090 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1092 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1094 struct inode *inode = file_inode(file);
1095 struct ll_sb_info *sbi = ll_i2sbi(inode);
1096 struct obd_ioctl_data *data;
1100 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1101 PFID(ll_inode2fid(inode)), inode, cmd);
1103 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1104 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1107 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1109 case FSFILT_IOC_GETFLAGS:
1110 case FSFILT_IOC_SETFLAGS:
1111 RETURN(ll_iocontrol(inode, file, cmd, arg));
1112 case FSFILT_IOC_GETVERSION_OLD:
1113 case FSFILT_IOC_GETVERSION:
1114 RETURN(put_user(inode->i_generation, (int __user *)arg));
1115 /* We need to special case any other ioctls we want to handle,
1116 * to send them to the MDS/OST as appropriate and to properly
1117 * network encode the arg field.
1118 case FSFILT_IOC_SETVERSION_OLD:
1119 case FSFILT_IOC_SETVERSION:
1121 case LL_IOC_GET_MDTIDX: {
1124 mdtidx = ll_get_mdt_idx(inode);
1128 if (put_user((int)mdtidx, (int __user *)arg))
1133 case IOC_MDC_LOOKUP: {
1134 int namelen, len = 0;
1138 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1143 filename = data->ioc_inlbuf1;
1144 namelen = strlen(filename);
1146 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1147 GOTO(out_free, rc = -EINVAL);
1150 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1152 CERROR("%s: lookup %.*s failed: rc = %d\n",
1153 ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1158 obd_ioctl_freedata(buf, len);
1161 case LL_IOC_LMV_SETSTRIPE: {
1162 struct lmv_user_md *lum;
1171 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1176 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1177 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1178 GOTO(lmv_out_free, rc = -EINVAL);
1180 filename = data->ioc_inlbuf1;
1181 namelen = data->ioc_inllen1;
1184 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1185 GOTO(lmv_out_free, rc = -EINVAL);
1187 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1188 lumlen = data->ioc_inllen2;
1190 if (lum->lum_magic != LMV_USER_MAGIC ||
1191 lumlen != sizeof(*lum)) {
1192 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1193 filename, lum->lum_magic, lumlen, -EFAULT);
1194 GOTO(lmv_out_free, rc = -EINVAL);
1197 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1198 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1200 mode = data->ioc_type;
1202 rc = ll_dir_setdirstripe(inode, lum, filename, mode);
1204 obd_ioctl_freedata(buf, len);
1208 case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1209 struct lmv_user_md lum;
1210 struct lmv_user_md __user *ulump =
1211 (struct lmv_user_md __user *)arg;
1214 if (copy_from_user(&lum, ulump, sizeof(lum)))
1217 if (lum.lum_magic != LMV_USER_MAGIC)
1220 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1224 case LL_IOC_LOV_SETSTRIPE: {
1225 struct lov_user_md_v3 lumv3;
1226 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1227 struct lov_user_md_v1 __user *lumv1p =
1228 (struct lov_user_md_v1 __user *)arg;
1229 struct lov_user_md_v3 __user *lumv3p =
1230 (struct lov_user_md_v3 __user *)arg;
1232 int set_default = 0;
1234 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1235 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1236 sizeof(lumv3p->lmm_objects[0]));
1237 /* first try with v1 which is smaller than v3 */
1238 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1241 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1242 if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1246 if (inode->i_sb->s_root == file_dentry(file))
1249 /* in v1 and v3 cases lumv1 points to data */
1250 rc = ll_dir_setstripe(inode, lumv1, set_default);
1254 case LL_IOC_LMV_GETSTRIPE: {
1255 struct lmv_user_md __user *ulmv =
1256 (struct lmv_user_md __user *)arg;
1257 struct lmv_user_md lum;
1258 struct ptlrpc_request *request = NULL;
1259 union lmv_mds_md *lmm = NULL;
1262 struct lmv_user_md *tmp = NULL;
1266 int max_stripe_count;
1270 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1273 max_stripe_count = lum.lum_stripe_count;
1274 /* lum_magic will indicate which stripe the ioctl will like
1275 * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1276 * is for default LMV stripe */
1277 if (lum.lum_magic == LMV_MAGIC_V1)
1278 valid |= OBD_MD_MEA;
1279 else if (lum.lum_magic == LMV_USER_MAGIC)
1280 valid |= OBD_MD_DEFAULT_MEA;
1284 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1287 GOTO(finish_req, rc);
1289 /* Get default LMV EA */
1290 if (lum.lum_magic == LMV_USER_MAGIC) {
1291 if (lmmsize > sizeof(*ulmv))
1292 GOTO(finish_req, rc = -EINVAL);
1294 if (copy_to_user(ulmv, lmm, lmmsize))
1295 GOTO(finish_req, rc = -EFAULT);
1297 GOTO(finish_req, rc);
1300 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1301 if (max_stripe_count < stripe_count) {
1302 lum.lum_stripe_count = stripe_count;
1303 if (copy_to_user(ulmv, &lum, sizeof(lum)))
1304 GOTO(finish_req, rc = -EFAULT);
1305 GOTO(finish_req, rc = -E2BIG);
1308 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1309 OBD_ALLOC(tmp, lum_size);
1311 GOTO(finish_req, rc = -ENOMEM);
1313 mdt_index = ll_get_mdt_idx(inode);
1315 GOTO(out_tmp, rc = -ENOMEM);
1317 tmp->lum_magic = LMV_MAGIC_V1;
1318 tmp->lum_stripe_count = 0;
1319 tmp->lum_stripe_offset = mdt_index;
1320 for (i = 0; i < stripe_count; i++) {
1323 fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1324 mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1326 GOTO(out_tmp, rc = mdt_index);
1328 tmp->lum_objects[i].lum_mds = mdt_index;
1329 tmp->lum_objects[i].lum_fid = fid;
1330 tmp->lum_stripe_count++;
1333 if (copy_to_user(ulmv, tmp, lum_size))
1334 GOTO(out_tmp, rc = -EFAULT);
1336 OBD_FREE(tmp, lum_size);
1338 ptlrpc_req_finished(request);
1342 case LL_IOC_REMOVE_ENTRY: {
1343 char *filename = NULL;
1347 /* Here is a little hack to avoid sending REINT_RMENTRY to
1348 * unsupported server, which might crash the server(LU-2730),
1349 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1350 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1351 * server will support REINT_RMENTRY XXX*/
1352 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1355 filename = ll_getname((const char __user *)arg);
1356 if (IS_ERR(filename))
1357 RETURN(PTR_ERR(filename));
1359 namelen = strlen(filename);
1361 GOTO(out_rmdir, rc = -EINVAL);
1363 rc = ll_rmdir_entry(inode, filename, namelen);
1366 ll_putname(filename);
1369 case LL_IOC_LOV_SWAP_LAYOUTS:
1371 case IOC_OBD_STATFS:
1372 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1373 case LL_IOC_LOV_GETSTRIPE:
1374 case LL_IOC_MDC_GETINFO:
1375 case IOC_MDC_GETFILEINFO:
1376 case IOC_MDC_GETFILESTRIPE: {
1377 struct ptlrpc_request *request = NULL;
1378 struct lov_user_md __user *lump;
1379 struct lov_mds_md *lmm = NULL;
1380 struct mdt_body *body;
1381 char *filename = NULL;
1384 if (cmd == IOC_MDC_GETFILEINFO ||
1385 cmd == IOC_MDC_GETFILESTRIPE) {
1386 filename = ll_getname((const char __user *)arg);
1387 if (IS_ERR(filename))
1388 RETURN(PTR_ERR(filename));
1390 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1391 &lmmsize, &request);
1393 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1398 body = req_capsule_server_get(&request->rq_pill,
1400 LASSERT(body != NULL);
1406 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1407 cmd == LL_IOC_MDC_GETINFO))
1408 GOTO(skip_lmm, rc = 0);
1413 if (cmd == IOC_MDC_GETFILESTRIPE ||
1414 cmd == LL_IOC_LOV_GETSTRIPE) {
1415 lump = (struct lov_user_md __user *)arg;
1417 struct lov_user_mds_data __user *lmdp;
1418 lmdp = (struct lov_user_mds_data __user *)arg;
1419 lump = &lmdp->lmd_lmm;
1421 if (copy_to_user(lump, lmm, lmmsize)) {
1422 if (copy_to_user(lump, lmm, sizeof(*lump)))
1423 GOTO(out_req, rc = -EFAULT);
1427 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1428 struct lov_user_mds_data __user *lmdp;
1431 st.st_dev = inode->i_sb->s_dev;
1432 st.st_mode = body->mbo_mode;
1433 st.st_nlink = body->mbo_nlink;
1434 st.st_uid = body->mbo_uid;
1435 st.st_gid = body->mbo_gid;
1436 st.st_rdev = body->mbo_rdev;
1437 st.st_size = body->mbo_size;
1438 st.st_blksize = PAGE_CACHE_SIZE;
1439 st.st_blocks = body->mbo_blocks;
1440 st.st_atime = body->mbo_atime;
1441 st.st_mtime = body->mbo_mtime;
1442 st.st_ctime = body->mbo_ctime;
1443 st.st_ino = cl_fid_build_ino(&body->mbo_fid1,
1447 lmdp = (struct lov_user_mds_data __user *)arg;
1448 if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1449 GOTO(out_req, rc = -EFAULT);
1454 ptlrpc_req_finished(request);
1456 ll_putname(filename);
1459 case OBD_IOC_QUOTACTL: {
1460 struct if_quotactl *qctl;
1462 OBD_ALLOC_PTR(qctl);
1466 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1467 GOTO(out_quotactl, rc = -EFAULT);
1469 rc = quotactl_ioctl(sbi, qctl);
1472 copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1479 case OBD_IOC_GETDTNAME:
1480 case OBD_IOC_GETMDNAME:
1481 RETURN(ll_get_obd_name(inode, cmd, arg));
1482 case LL_IOC_FLUSHCTX:
1483 RETURN(ll_flush_ctx(inode));
1484 case LL_IOC_GETOBDCOUNT: {
1486 struct obd_export *exp;
1488 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1491 /* get ost count when count is zero, get mdt count otherwise */
1492 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1493 vallen = sizeof(count);
1494 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1495 KEY_TGT_COUNT, &vallen, &count);
1497 CERROR("get target count failed: %d\n", rc);
1501 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1506 case LL_IOC_PATH2FID:
1507 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1508 sizeof(struct lu_fid)))
1511 case LL_IOC_GET_CONNECT_FLAGS: {
1512 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1513 (void __user *)arg));
1515 case OBD_IOC_CHANGELOG_SEND:
1516 case OBD_IOC_CHANGELOG_CLEAR:
1517 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1520 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1521 sizeof(struct ioc_changelog));
1523 case OBD_IOC_FID2PATH:
1524 RETURN(ll_fid2path(inode, (void __user *)arg));
1525 case LL_IOC_GETPARENT:
1526 RETURN(ll_getparent(file, (void __user *)arg));
1527 case LL_IOC_FID2MDTIDX: {
1528 struct obd_export *exp = ll_i2mdexp(inode);
1532 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1536 /* Call mdc_iocontrol */
1537 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1538 (__u32 __user *)&index);
1544 case LL_IOC_HSM_REQUEST: {
1545 struct hsm_user_request *hur;
1552 /* We don't know the true size yet; copy the fixed-size part */
1553 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1558 /* Compute the whole struct size */
1559 totalsize = hur_len(hur);
1564 /* Final size will be more than double totalsize */
1565 if (totalsize >= MDS_MAXREQSIZE / 3)
1568 OBD_ALLOC_LARGE(hur, totalsize);
1572 /* Copy the whole struct */
1573 if (copy_from_user(hur, (void __user *)arg, totalsize))
1574 GOTO(out_hur, rc = -EFAULT);
1576 if (hur->hur_request.hr_action == HUA_RELEASE) {
1577 const struct lu_fid *fid;
1581 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1582 fid = &hur->hur_user_item[i].hui_fid;
1583 f = search_inode_for_lustre(inode->i_sb, fid);
1589 rc = ll_hsm_release(f);
1595 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1600 OBD_FREE_LARGE(hur, totalsize);
1604 case LL_IOC_HSM_PROGRESS: {
1605 struct hsm_progress_kernel hpk;
1606 struct hsm_progress hp;
1608 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1611 hpk.hpk_fid = hp.hp_fid;
1612 hpk.hpk_cookie = hp.hp_cookie;
1613 hpk.hpk_extent = hp.hp_extent;
1614 hpk.hpk_flags = hp.hp_flags;
1615 hpk.hpk_errval = hp.hp_errval;
1616 hpk.hpk_data_version = 0;
1618 /* File may not exist in Lustre; all progress
1619 * reported to Lustre root */
1620 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1624 case LL_IOC_HSM_CT_START:
1625 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1628 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1629 sizeof(struct lustre_kernelcomm));
1632 case LL_IOC_HSM_COPY_START: {
1633 struct hsm_copy *copy;
1636 OBD_ALLOC_PTR(copy);
1639 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1644 rc = ll_ioc_copy_start(inode->i_sb, copy);
1645 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1651 case LL_IOC_HSM_COPY_END: {
1652 struct hsm_copy *copy;
1655 OBD_ALLOC_PTR(copy);
1658 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1663 rc = ll_ioc_copy_end(inode->i_sb, copy);
1664 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1670 case LL_IOC_MIGRATE: {
1672 const char *filename;
1678 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1682 data = (struct obd_ioctl_data *)buf;
1683 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1684 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1685 GOTO(migrate_free, rc = -EINVAL);
1687 filename = data->ioc_inlbuf1;
1688 namelen = data->ioc_inllen1;
1689 /* \0 is packed at the end of filename */
1690 if (namelen < 1 || namelen != strlen(filename) + 1)
1691 GOTO(migrate_free, rc = -EINVAL);
1693 if (data->ioc_inllen2 != sizeof(mdtidx))
1694 GOTO(migrate_free, rc = -EINVAL);
1695 mdtidx = *(int *)data->ioc_inlbuf2;
1697 rc = ll_migrate(inode, file, mdtidx, filename, namelen - 1);
1699 obd_ioctl_freedata(buf, len);
1704 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1705 (void __user *)arg));
1709 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1711 struct inode *inode = file->f_mapping->host;
1712 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1713 struct ll_sb_info *sbi = ll_i2sbi(inode);
1714 int api32 = ll_need_32bit_api(sbi);
1715 loff_t ret = -EINVAL;
1718 mutex_lock(&inode->i_mutex);
1723 offset += file->f_pos;
1729 offset += LL_DIR_END_OFF_32BIT;
1731 offset += LL_DIR_END_OFF;
1738 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1739 (!api32 && offset <= LL_DIR_END_OFF))) {
1740 if (offset != file->f_pos) {
1741 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1742 (!api32 && offset == LL_DIR_END_OFF))
1743 fd->lfd_pos = MDS_DIR_END_OFF;
1744 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1745 fd->lfd_pos = offset << 32;
1747 fd->lfd_pos = offset;
1748 file->f_pos = offset;
1749 file->f_version = 0;
1756 mutex_unlock(&inode->i_mutex);
1760 static int ll_dir_open(struct inode *inode, struct file *file)
1763 RETURN(ll_file_open(inode, file));
1766 static int ll_dir_release(struct inode *inode, struct file *file)
1769 RETURN(ll_file_release(inode, file));
1772 const struct file_operations ll_dir_operations = {
1773 .llseek = ll_dir_seek,
1774 .open = ll_dir_open,
1775 .release = ll_dir_release,
1776 .read = generic_read_dir,
1777 #ifdef HAVE_DIR_CONTEXT
1778 .iterate = ll_iterate,
1780 .readdir = ll_readdir,
1782 .unlocked_ioctl = ll_dir_ioctl,